← Om Badgujar
Project / 03
Computer Vision · Group Project

AI Fitness Monitor

A real-time AI fitness monitoring app built with a team of four. MediaPipe extracts pose landmarks, a classifier scores squat form, and a Flutter client streams frames through a FastAPI backend for live feedback.

§ Problem

Why it exists.

Home workouts are effective only if the form is right — and there's no coach watching. We wanted a phone-first tool that could grade a squat in real time without shipping a wearable.

§ Approach

How it works.

  1. 01Owned the ML pipeline: pose landmark extraction with MediaPipe, feature engineering on joint angles, squat-posture classification with scikit-learn.
  2. 02Wrapped the model in a FastAPI service exposing a low-latency inference endpoint.
  3. 03Integrated the backend with a Flutter mobile app via REST for real-time posture prediction and feedback.
  4. 04Used Firebase for auth and session storage so users could track progress across workouts.
§ Key features

What it does.

  • Real-time pose landmark extraction with MediaPipe on live camera frames.
  • Squat-form classifier trained on engineered joint-angle features.
  • FastAPI inference endpoint consumed by a Flutter mobile client.
  • Firebase auth + session storage for tracking workouts over time.
§ Tech stack

Built with.

PythonFastAPIFlutterMediaPipeOpenCVScikit-learnFirebase
  • Real-time squat feedback on-device via the mobile app.
  • Clean separation of ML backend and Flutter client — either side can iterate independently.
  • Shipped as a four-person group project with clear module ownership.